Max-Diff Specifications

From Q
Jump to: navigation, search

In general, the best way to set up max-diff in Q is via the Online Library, using Marketing - Max-Diff - Max-Diff Setup from an Experimental Design.

This page discusses the underlying details about how the data needs to be setup (i.e., the QScript in the Online Library sets up the data according to these specifications).

Max-Diff experiments are questions that present respondents with a series of options and ask them which is best and which is worst. For example, a very simple max-diff experiment may consist of the following two tasks:

Q5a.   Which of these companies do you like the most and which do you like the least?  

Like 			Like 
most			least
○	Apple		○
○	Microsoft	○
○	IBM		○
○	Google		○

Q5b.   Which of these companies do you like the most and which do you like the least? 

Like 			Like 
most			least
○	IBM		○
○	Toshiba		○
○	Dell		○
○	Apple		○

Max-diff questions are setup as follows:

  1. The number of variables required is AT where A is the total number of alternatives (e.g., the total number of brands, or, the total number of attributes being evaluated) and T is the number of tasks. In the example shown above, T = 2 and A = 6 (i.e. even though only four brands appear in each task, the total number of brands appearing in the two tasks is six), so 12 variables are required in total.
  2. The first A variables relate to the first task, variables A+1 to 2A relate to the second task, etc.
  3. The Variable Name for each variable in the data file should be unique. It is useful to create informative variable names, such as: Task1Alt1, Task1Alt2, Task1Alt3, Task1Alt4, Task1Alt5, Task1Alt6, Task2Alt1, Task2Alt2, Task2Alt3, Task2Alt4, Task2Alt5 and Task2Alt6.
  4. The Variable Label for each variable should be the description of the option, even if the option was not presented. For example, for the two tasks, labels would be: Apple, Microsoft, IBM, Google, Dell, Toshiba, Apple, Microsoft, IBM, Google, Dell and Toshiba. Note that:
  • The variable labels are repeated the same number of times as there are tasks. For example, with four alternatives, called A, B, C and D, and six tasks, the labels would be: A B C D A B C D A B C D A B C D A B C D A B C D. You can enter the labels for just the first task and then use Copy and Paste in Q to replicate the labels for the remaining tasks.
  • The variable labels are always in the same order.
  • No additional information is contained in the variable labels (e.g., no question name, nothing indicating the task).
For example, the variable names and corresponding labels for the questions shown previously could be laid out like this (a different ordering of the brands could be used):
Task1Alt1	Task1Alt2	Task1Alt3	Task1Alt4	Task1Alt5	Task1Alt6	Task2Alt1	Task2Alt2	Task2Alt3	Task2Alt4	Task2Alt5	Task2Alt6
Apple		Microsoft	IBM		Google		Dell		Toshiba		Apple		Microsoft	IBM		Google		Dell		Toshiba
  1. Each variable should contain one of four possible values for each respondent:
  • A NaN if the option was not shown.
  • A 1 if the option was chosen as best.
  • A -1 if the option was chosen as worst.
  • A 0 if the option was shown but not selected.
With the example, if the options are laid out in the following order: Apple, Microsoft, IBM, Google, Dell and Toshiba, then the data should look something like this:
Task1Alt1	Task1Alt2	Task1Alt3	Task1Alt4	Task1Alt5	Task1Alt6	Task2Alt1	Task2Alt2	Task2Alt3	Task2Alt4	Task2Alt5	Task2Alt6
0		1		-1		0		NaN		NaN		1		NaN		-1		NaN		0		0
0		1		-1		0		NaN		NaN		0		NaN		-1		NaN		1		0
1		0		-1		0		NaN		NaN		1		NaN		-1		NaN		0		0
0		0		1		-1		NaN		NaN		1		NaN		0		NaN		0		1
In this example, in the first task the respondent chose Microsoft as Best and IBM as worst, while in the second task they chose Apple as Best and IBM as worst again.

See also

Personal tools
Namespaces

Variants
Actions
Navigation
Categories
Toolbox